IMPROVING DETECTION OF SOYBEAN SCN INFESTATION USING MULTI-SCALE REMOTE SENSING AND A NOVEL VEGETATION INDEX

Author

Yuhua Wang, Southern Illinois University CarbondaleFollow

Date of Award

5-1-2025

Degree Name

Master of Science

Department

Geography and Environmental Resources

First Advisor

Li, Ruopu

Abstract

Soybean Cyst Nematode (SCN) is a pathogen with serious impacts on soybean yields. Traditional field-based assessment is labor-intensive and often ineffective for early interventions and the existing spectral vegetation indices (VIs) from remotely sensed data also lack the ability of accurately detect SCN infested plants. In this study, a greenhouse-based experiment was designed to collect a total of 100 hyperspectral data sets from 20 soybean plants from the 68th to 97th day after planting. Through statistical analysis, feature selection, and classification comparison of the hyperspectral data using seven classifiers, a new spectral VI, called SCNVI, was proposed based on the selected bands 338 nm and 665 nm. Three plant stress categories were defined based on initial egg inoculation levels: healthy (0 egg), moderate stress (1000 or 5000 eggs) and severe stress (10,000 eggs). The results showed that compared with the healthy plants, stressed plants significantly increased spectral reflectance in both UV and visible regions. ANOVA analysis indicated statistically significant differences in spectral reflectance among the three defined stress levels: healthy, moderate stress, and severe stress. Based on the significant bands identified through ANOVA (p < 0.05), a Principal Component Analysis (PCA) was conducted, which showed that the first two components (PC1 and PC2) captured 97.18% of the total variance. Moreover, seven classifiers led to their top 10 bands selected with most of them falling in the region from 511 nm to 672 nm with several in the UV and red-edge region, such as 338 nm and 699 nm. Integrating each of the top 10 bands with seven classifiers resulted in an accuracy of 70% for distinguishing between healthy and stressed plants, but the accuracies of 40% to 60% for three-class classification. The SCNVI, coupled with eXtreme Gradient Boosting (XGBoost), achieved an accurate classification of 70% for three classes, and significantly outperformed the 13 traditional VIs by increasing the accuracy by more than 67%. Therefore, integrating the SCNVI and XGBoost algorithm provided great potential of improving detection of SCN infestation for soybean lands.To further evaluate the applicability of the proposed index in real-field conditions, a UAV-adapted version of the SCNVI (SCNVI_UAV) was developed for use with multispectral imagery, since typical UAV sensors do not capture the UV spectrum. In this adaptation, an alternative band 570 nm was used to approximate the sensitivity normally provided by the UV band. The UAV-derived SCNVI time series data were clustered into three groups representing healthy, moderate stress and severe stress levels. Statistical validation showed a strong correlation between the SCNVI_UAV-based clusters and field-observed SCN egg counts, with significant differences between clusters (p < 0.05). Overall, integrating the hyperspectral-derived SCNVI with advanced machine learning techniques and adapting the index for UAV platforms provides a promising foundation for the early detection and management of SCN infestations, ultimately contributing to more sustainable soybean production.